Data Mining Methods For Marketing Intelligence

With the development of Internet technologies, marketing has become more and more intelligent. Social network, which is based on Internet technologies, has become an important platform for marketing, attracting attention from both industry and a-cademia. On the other side, Internet has made information more transparent and huge amount of data available for analysis and mining, which can be used to support mak-ing smarter marketing decisions. In this paper, we research into viral marketing and real estate marketing, and explore two important characteristics in marketing, name-ly, the targeted customers and days on market. Specifically, our contributions can be summarized as follows.First, we find that, in traditional influence maximization models for viral market-ing, the category distribution of the influenced crowd is often imbalanced, which indi-cates the lack of diversity of the influenced crowd. Therefore, we propose the frame-work of diversified influence maximization. We propose a class of diversity measures, and by applying any of these measures to the framework, we could obtain a specif-ic optimization objective. We prove that the optimization objective is non-decreasing and submodular. Considering that computing the influence distribution of each node is time-consuming, we propose to further relax the original formulation. Namely, we use the diversity of the seed set instead of the diversity of the influenced crowd. Due to the homophily phenomenon, which is common in social networks, this relaxation is reasonable. We further integrate this idea into some widely used heuristics for influ-ence maximization, such as degree centrality and PageRank. Experimental results have verified that our method could achieve better diversity of the influenced crowd, and it is easy to obtain a trade-off between diversity and influence.Second, days on market (DOM) refers to the number of days a property is on the active market, which is an important measurement of market status in real estate indus-try. It is also important for the decision making of buyers and sellers, e.g. evaluating the popularity of a house. There has been a lot of research on analyzing and interpret-ing the relationship between price and DOM. However, many of these research works present very different results, making it a controversial topic, and no conclusion has been established. In this study, instead of trying to figure out the relationship between price and DOM, we focus on accurately predicting the DOM by employing machine learning techniques. Specifically, we first obtain a transaction data set in Beijing from a major commercial real estate agency in China, and some other data sets containing re-lated geographical information. Based on these data, we extract5categories of features, namely, house profile features, residential community features, geographical features, temporal features, and meta features. Then we propose a multi-task learning regression model based on Tobler’s First Law of Geography, and propose an algorithm to solve this problem. Finally, our experimental results have shown that our method is better than all the baselines on nMSE and MAE metrics.